Outline Working group goals B2hh selection: current status CP asymmetry fit: status and future plans Working group organization
Goal: do the full B hh exercise! Bhh channels have been identified by the PPG amongst the key measurements for the initial phase of LHCb The idea is that the physics book should be a path toward realistic analyses, that means study how these measurements will be performed by giving not only the expected statistical sensitivity, but also the list of systematics to consider and the actual procedures to tackle them; it also means going in as much details as possible/useful in the various calibrations and checks which will be needed, including control samples, etc... Bhh can be in fact an effective proof of principle of a realistic analysis, in particular crucial to understand and demonstrate the LHCb PID performance with hadrons Bhh can be used to practice the CP fit tools
B selection strategy Stable since (2004) LHCb Note 2003-123 Designed to maximize signal efficiency and minimize backgrounds Two major sources of backgrounds taken into account Combinatorial background from bb inclusive events Due to the huge minimum bias MC statistics required for a detailed study of combinatorics, we make the plausible assumption that most of the combinatorial background will come from beauty events (presence of high p T and large IP tracks from B) For the moment we prefer not to adopt dangerous tricks (e.g. cuts at generator level or fast MC simulations) Specific background from B decays with same two-track topology e.g. B d K + π -, B s K + K -, B s π + K - as backgrounds for B d π + π - Selection cuts are simultaneously optimized in order to maximize S/sqrt(S+B) After event selection, tagging and trigger algorithms are tuned on selected events
Selection cuts See V.Vagnoni talk at CP meeting 14 th July 2005 π +, K + Beam axis (z) IP B IP IP L B 0 (s) π, K Particle ID Each charged track identified as a Pion or Kaon using the Particle ID detectors (RICHs in particular) Reconstruction of long tracks a B flights 1 cm on average Tracks selection cuts DLL cut (PID) Max[pT (h + ), p T (h - )] Min[pT (h + ), p T (h - )] Max[IP/σIP (h + ), IP/σ IP (h - )] Min[IP/σIP (h + ), IP/σ IP (h - )] B Selection cuts pt IP/σIP L/σL Invariant mass χ 2 of common vertex
B selection (2005) Momentum cuts have been replaced by DLL cuts: Bd2PiPi: Combined pid (excl.) but vetoing pion hypotesis with DLL(K pi)> 2 for the Kaon hypotesis (default is DLL(K pi) > 2) Bd2KPi: Combined pid (excl.) Bs2KK: Combined pid (excl.) Bs2PiK: Combined pid (excl.) selection but pions must have DLL(pi K)> 2
S/B (used for Feb 04 studies) Event type Assumed BR (x10-6 ) N B /N S Untagged annual yield εd 2 B d π + π - 4.8 0.42 26000 4% B d K + π - 18.5 0.16 135000 4% B s K + K - 18.5 B s π + K - 4.8 0.31 0.67 37000 5300 6% 6% Annual yields after L0+L1 triggers and offline selection (assumed σ bb = 0.5 mb) N B /N S here quoted only from combinatorial bb background Presented at Joint meeting in Feb 04
Analysis strategy LHCb Note 2003-124 What do we want to do with selected Bs? Build the asymmetries Extract time and mass distributions for signal and background candidates using PID information (to be changed) Fit simultaneously mass and time distributions with an extended unbinned maximum likelihood fit Extract gamma from Asymmetry with a Bayesian approach
CP asymmetries Using U spin symmetry d = d' and ='. We end up with 3 unknowns and 4 equations
(old current) Fit strategy B d K + π is a flavour specific decay, hence it can be used to extract the wrong tag probability from data B d π + π (B s K + K ) and B d K + π (B s π + K ) have the same two track topology (the full simulation shows the same tagging power) In order to extract CP asymmetries and mistag fraction simultaneously from data a combined extended maximum likelihood fit of B d π + π and B d K + π (B s K + K and B s π + K ) event samples
(old current) Likelihood fit The likelihood fit is performed with 17 free parameters: Adir and A mix for B d π + π - (B s K + K - ) Charge asymmetry for Bd K + π - (B s π + K - ) Mean Bd (B s ) mass and mass resolution (2 parameters) 1 parameter for background mass distribution of Bd π + π - (B s K + K - ) 1 parameter for background mass distribution of Bd K + π - (B s π + K - ) 2 parameters for background proper time of Bd π + π - (B s K + K - ) 2 parameters for background proper time of Bd K + π - (B s π + K - ) M, Γ, Γ for the Bd (B s ) Mistag fraction for Bd π + π - and B d K + π - (B s K + K - and B s π + K - ) Tagged event yield for Bd π + π - (B s K + K - ) Tagged event yield for Bd K + π - (B s π + K - )
Fit input (I): Proper time and resolution Acceptance suppression at low proper time due to cuts on distance of flight and IP σ 1 =34±1 fs σ 2 =67±3 fs (15%) Asyntotically flat Resolution on proper time fitted with double gaussian
Fit input (II): Bkg proper time distribution Only a handful of bb events surviving after event selection due to limited (10 7 ) fully simulated bb event sample Exponential times acceptance function bb background lifetime 1.1 ps
Fit input (III): mass spectra Specific background only B d π + π - B d K + π - if not using RICH σ 17 MeV/c 2 B s K + K - B s π + K - Mostly B d K + π -
Fit input (IV): background mass distr. Only a handful of bb events surviving after event selection due to limited (10 7 ) fully simulated bb event sample Exponential fit
Calibration issues Particle ID for pions and kaons we have large D* samples Calibration for protons would be needed as well. Can we do this with Lambda decays? Invariant mass & Vertex resolution mass resolution is crucial for the 2 body studies. Large J/Psi and friends samples available All of this is fairly clear, but we need still to formalize how the information obtained from various light house processes is propagated to the CP fits
Large D* samples used for PID studies R. Muresan Plan to use D * D 0 π, D 0 Kπ: Golden kinematics easy to supress the bkg: (M D* M D 0)=144.5 MeV (M π =139.5 MeV) Good branching ratios: BR( D * D 0 π ) = 67.7 %, BR( D0 K π ) = 3.83 % All D 0 s, true D 0 s Selected events (HLTFilterDstars) After L0, L1 6768 + HLT 2099 (31%) + off line HLT 2718 (40%) + HLT & off line HLT 1565 (23% )
Results (after HLT) K, Pi RICH PID decision Bd >D*π L0, L1, HLT flag, off line HLT cuts R. Muresan Those events can be used to calibrate the PID (see few slides later) kaons pions pions 1584 k: 1571 k 7 are 1 is a 5 bkg 1584 : 1572 2 are k 4 are 5 bkg Entries below 0 4 ghosts,1 muon Entries below 0 2 ghosts one kaon Entries below 0 1 ghost, 2muons,1 el
Fit technology: RooFit Started with a Fortran code (using Minuit) and now ended with a RooFit based fit: results are consistent :)!! Should push for a common code for all the other analyses that are doing such likelihood fits... (see proposal in CP meeting of 14 Jul 05) The PID information is not used, now, in the fit: plans are to go for a simultaneous fit of all the 8 different B to hh samples. Can be done! (G. Raven CP meeting)
Fitting using the PID information 8 signal samples: B 0,B s pi + pi,k + pi,pi + K,K + K ( Convention: positive charge first) The above 8 samples, for non perfect PID, crossfeed amongst each other Eg. Bs Kpi is a background for B0 pipi, and vice versa Need to relative Kpi yield for pipi background estimate, but need pipi yield to get Kpi background Separation of samples depends on PID performance, which will depend on eg. track momenta, which eg. distorts the shape of Kpi cross feed background in mpipi Cutting on PID to select 8 samples will result in non trivial analysis Invariant mass also has some power to separate the signals.. Alternative: use both invariant mass and PID observables directly in fit, get all 8 signals simultaneously Let MINUIT worry about correlations But need to use consistent observables: For Kpi, should compute invariant mass under pipi hypothesis to allow comparison with pipi
Fitting using the PID information (II) Need to Model the m(pipi) mass. Assume M(π + π ) is a Gaussian, with a mean which depends on the observed value of β (Similar to resolution models which depend on the observed perevent propertime error derived from the vertex and momentum) Use conditional PDF to describe this distribution: L(m,) = L(m) L() Model the PID Need to parameterize the momentum (and charge) dependence of PID observables RICH TDR shows this plot that is just what you want This would lead to an observable PID with the following properties L(PID,p) = L(PID p)l(p) Again, this is a Conditional PDF, given the momentum distribution of a track.
Future Plans PID observables PID categorization and simultaneous fit to all B to hh samples HLT Trigger studies Tagging studies () D to hh studies (helping the PID categorization)
Conclusions and outlook Started a joint work on B >hh in order to perform the full exercise in time for Phys Book (or whatever else)